Rapid reconfiguration in an acousto-optic crossbar interconnection network

Rapid reconfiguration in an acousto-optic crossbar interconnection network
Rapid Reconfiguration in an Acousto-Optic Crossbar
Interconnection Network
J. GamoArandaa, R. Mcleodb, P. R. Horchea and K. Wagnerc
aDept. de TFO, ETSI Telecomunicación, Universidad Politécnica de Madrid, Spain
bETEK Dynamics, 1865 Lundy Ave., San Jose CA, 95131
cDept of ECE/OCS, University of Colorado, Boulder CO, 80309
We demonstrate the operation and rapid reconfiguration of a 12x12 Acousto-Optic Photonic Crossbar (AOPC). This
AOPC can implement any desired permutation, fan-in, or fan-out interconnection between any subset out of twelve
single-mode input fibers into any subset out of twelve single-mode output fibers. The system uses one large-aperture
Acousto-Optic Deflector (AOD) driven by a sum-of-tones RF-waveform produced by an arbitrary waveform generator
( ARB) and computed from an experimentally measured lookup table, thus reducing the control complexity of the
system. The design, based on the momentum-space technique, includes optical and acoustical rotation for the AOD,
in order to optimize the efficiency of the desired interconnections and minimize the undesirable negative first-order
acoustooptic Bragg-diffractions. A limitation in this type of systems is the unavoidable reconfiguration (dead) time
introduced by the AOD itself, which can result in crosstalk between the individual input channels during that period
oftime. In this paper, we experimentally investigate the reconfiguration time ofthis AOPC, by switching between two
different crossbar patterns, and then measuring the time during which the detected signal can not be individually
resolved for each input channel. Coupling efficiency problems and alignment procedures are also discussed and
Keywords: Acoustooptic devices, photonic switch, optical crossbar
Fig. 1 shows a schematic layout of the Acousto-Optic Photonic Crossbar (AOPC). The complete design equations
for multi-channel optical crossbar implemented in a single-transducer AO crystal have previously been presented.'
A 12x12 version of such a switch has been built2 •3 This AOPC can simultaneously and independently deflects each
of the twelve input fibers to any or all of the 12 output fibers by means of a RF multi-tone driving signal applied
to the Bragg cell. The individual frequencies are taken from an experimentally-calculated lookup table and then
combined together using an ARB and real-time experiment.
Single—mode Short focal length
Collimation Lens
Long focal length
Focusing Lens
Figure 1 .
J . Gamo-Aranda:
Schematic layout of the acoustooptic crossbar switch
E-mail: jgamotfo.upm.es
K. Wagner: E-mail: kelvin©colorado.edu
Part of the SPIE Conference on Photonic Devices and Algorithms for Computing
Denver, Colorado S July 1999 SPIE Vol. 3805 . 0277-786X/99/$1O.OO
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
k—vectors of
" \of acousti
k—vectors of
ordinary inputs
Kord ,'
K (1/rn) in [001]
RF Frequency
Figure 2. a) Simplified momentum-space of the acousto-optic interaction for a 5x5 switch in an anisotropic crystal
such as Te02. Note that the inputs are widely spaced and tangentially matched, while the outputs are closely spaced
and not tangentially matched. The resulting bandshapes, shown in b) are thus unusually narrow and arranged to
be non-overlapping. The 25 individual frequencies, consisting of 5 within each band (diamonds), when appropriately
chosen, uniquely deflect one input to one output
Fig. 2 a shows a simplified view of the momentum-space diagrams and analysis employed in the design of the
AOPC. The system operates in a novel regime referred to as anti-tangential, where the input momentum surface is
tangent to the locus of the acoustic wave launched by the transducer. This mode of operation allows each input plane
wave to be Bragg-matched only to a narrow range of acoustic frequencies, which is exactly the condition required in
a low-loss switch.
This AOPC was designed to operate at a laser wavelength of 850 nm for demonstration purposes. For longdistance communication systems, the optical wavelength should be chosen in the low-loss windows at 1300 or 1550
nm, but the physical and operating principle of the switch are not dependent on optical wavelength. Therefore, the
850 nm version is adequate to experimentally verify the performance of the switch.
An important limitation in these type of systems is the unavoidable reconfiguration (dead) time. During the
propagation of the reconfiguration transient the system exhibits crosstalk between the individual input channels,
when switching between two RF multifrequency waveforms, due to the acoustic transit time across the illuminated
aperture of the AOD. In using such a system for data transmission applications it is necessary to measure this
dead time, so that it can be accounted to, or compensated for (for instance using latency hiding techniques4) in the
network control.
The rest of this paper describes the experimental implementation of this 12x12 AOPC and measurements of its
rapid reconfiguration. Section 2 briefly describes some optical-design aspects involved in the AOPC implementation.
Hardware and software instrumentation is described in Section 3. In section 4, experimental results regarding the
reconfiguration-time measurements of the switch are presented, along with a description of the alignment difficulties
encountered and how we solved them. Finally, Section 5 summarizes our findings.
The design step used momentum-space based Fourier optics techniques, to analyze the diffraction of an initially
Gaussian beam when propagating through a 4-f optical system with a non-ideal Bragg cell.1 By optically and
acoustically rotating the AO cell (a Te02 crystal) both central-frequency tuning and Bragg-mismatching of the
symmetrical negative first-order defiections were achieved.
Single-mode (at 850 nm) patch-cords (5 jim core, 125 1um cladding) were used in both input and output planes
of the switch. The input and output spacing in the V-groove arrays were 750 jim and 250 jim, respectively, but
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
Figure 3. Lens folding by combining a positive and a negative lens.
the output spots were large and just resolved, and were focused into the fibers with a matched 250 m pitch lenslet
array. The first order design2 using OSLO© (a commercial package for optical system design5) revealed that the
required optical path length for the system shown in Fig. 1 was 14 m long.
Since 14 m length optical systems are inconvenient, a lens folding technique utilizing corrected aplanatic lenses
was employed instead. The use of a properly spaced positive and negative lens can implement a long focal length
in a short optical path.6 The consequence of the unwanted stretching of the front focal length is a slight deviation
from telecentric operation, but still within the acceptance cone of the fibers.
Since the large-aperture Bragg cell has a rectangular (non-square) shape, the incoming optical beam must be
anamorphically compressed before the cell, and then uncompressed back to its original circular shape after passing
through the cell. While cylindrical-lens telescopes might be considered, the aberration limited performance are
unacceptable. The same solution can be achieved by using two identical compressor-prism pairs. Fig. 4 shows the
operation of one of these prism pairs. The magnification of the pair:
M (cos 012 cosO22"
\ cos0 cos 021)
is only dependent upon the angles of refraction 023, calculated from Snells law. In our 12x12 AOPC, the required
magnification to properly fit the 45 mm wide beam to the 6.5 mm high transducer is 1/7. The two prism pairs
are mounted on similar special-purpose mounts which provide coarse control over six degrees of freedom and fine
adjustment of the two critical angles (tilt about the optical axis and rotation in the plane of Fig. 4).
To couple the just-resolved scanned beams (80 jim beam width, 250 tm spacing) into the over-resolved fibers (5
m core diameter, 250 m spacing), a 1-D lenslet array is used. This lenslet array is mounted on a precision five-axis
stage, while the output fiber V-groove is positioned on a motorized version of a similar stage.
Fig. 5 a shows a schematic outline of the switch testbed components. The entire system is controlled by a PC
running LABVIEW© , with the top level control panel interface shown in Fig. 5 (which controls and monitors the
different tasks involved in the operation of the switch). The system can be divided into the following four functional
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
Figure 4. Performing of the anarmorphical prism-pair compressor.
3.1. Input-generation subsystem
The optical input to the switch is provided by twelve laser diodes operating at 850 nm Each of the laser diodes
(Spectra Diode Labs© Mod. 5401, packaged and fiber-pigtailed by Seastar Optics©) is independently modulated
at 100 Mb/s. This 1.2 Gb/s of data is supplied by one of the two channels of a LeCroy© 9109 Arbitrary Function
Generator (ARB) , while the other channel provides the analog RF switching signal. Data patterns as well as laser
parameters (temperature, average optical power, bias current, and modulation depth) for each channel are fully
controlled by the user through the corresponding software modules.
3.2. Output-detection subsystem
The switched optical signals are detected using twelve fiber-coupled, amplified, high-speed avalanche photo-diodes
(APDs) detectors (EG&G© Mod. C30657-250). These APDs have a 250 MHz bandwidth, 0.090 pW/Hz'/2 noise
equivalent power and .33 V/jW responsivity, and are supported by custom circuits which provide both high-speed
and average voltage outputs. The high-speed output can be monitored with an oscilloscope, while the average ouput
signals are fed to a multi-channel analog-to-digital (A/D) converter on a multi-purpose I/O board, and are then read
and displayed by the corresponding software panel on the controller which allows system performance monitoring
and feedback control. The APDs are protected with a current limiting resistor and in addition, if one or several
of the detected average signals are above a predefined threshold, the control software sends a cut-off signal to the
high-voltage APD power supplies.
3.3. Alignment control
The use of single-mode fibers imposes severe alignment restrictions on this system. Commercially-available devices
can align single-channel based systems, but in this switch as many as twelve optical channels must be simultaneously
aligned. For this purpose, the multi-channel, average optical power monitoring hardware and software described
in Section 3.2 can be used. Since the five-axis NewFocus© Picomotor stage carrying the array of output fibers
can be driven under computer control, a software module was built to perform the switch auto-alignment. This
software module calls the average optical power software and tries to maximize the detected signal by applying small
excursions to the Picomotor actuators. Unfortunately, the non-ideal performance of the Picomotor hardware makes
multi-dimensional searches difficult, therefore a more pedestrian approach is described in Section 4.1.
3.4. Switching control
To configure the switch, the controller must select and sum several RF tones, one for each desired crossbar-connection
point. These individual frequencies are chosen with random phases from a 144 lookup frequency table (which must
be measured experimentally in a previous step), calculated as 8 bits waveform sequences, tested for peak power
limitations (and, if they fail, the phases are re-randomized until they pass), and finally sent to the LeCroy© ARB
(along with the desired interconnects pattern and data streams for the lasers). The 100 MHz sampling rate of the
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
Figure 5. a) Layout of the data/control lines for the components of the photonic switching testbed. b) The top
level Labview GUI interface allows access to the control panels for each of the system components.
ARB provides nearly 50 MHz of Nyquist limited bandwidth, which is mixed with approximately 55 MHz RF tone
to provide the 55-105 MHz operating frequencies for the optically rotated Te02 AOD. The output of the mixer is
amplified and applied to the AOD to control the switch configuration.
4.1. Alignment procedures
A "central interconnects" (input #6 to output #6) was used in aligning the system. A collimation test for the
diffracted beam using a shear plate after the beam is shrunk by the first prism pair and re-expanded by the second
prism pair is shown in Fig. 6. The parallelism of the fringes on the shear plate indicates a well-collimated and
diffraction-limited beam.
The procedure for aligning the diffracted beams (focussed into 80 im spots by the output folded Fourier lens)
onto the lenslet array, and aligning the lenslet array to the fiber cores in the V-grooves is explained here. Fig. 8
shows this procedure. A 5 jm thick pellicle beam splitter (BS) is placed before the lenslet array, and two CCD
cameras are used: CCD#1 images the lenslet array through the BS, while CCD#2 is placed in a plane conjugate
to the lenslets and detects the reflected spot coming from the BS. To properly set CCD#2 at the required distance,
the lenslet array is first imaged by CCD#1, then CCD#2 is illuminated and displaced along the OX axis until its
sensor array is imaged by CCD#1. At that position, both the lenslet array and CCD#2 are at the same distance
from the BS.
Fig. 7 shows an image of the output lenslet array using CCD#1. To facilitate the alignment, light was backpropagated from three output fibers in the V-groove linear array (note the three light spots, corresponding to output
fibers #1, #6 and #12). The V-groove fiber array is properly positioned with respect to the lenslets when the three
spots remain vertically aligned on the TV monitor while moving CCD#1 back and forth along OX axis. The size of
these spots is used to set the distance s between the lenslet and the V-groove planes.
Finally the folded focusing lenses Li and L2 are set to their correct positions, by launching light backwards and
ensuring that it is collimated, and by measuring the focal length.
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
S: *
4 *4*•••
*0 *4 * * 4 *
Figure 6. Beam-collimation testing after collimation, passing through the first prism pair,
Figure 7. Image of the output lenslet array
showing the three light spots, corresponding
to output fibers #1, #6 and #12 (from left to
being deflected by the AOD, and recircularized
by the second prism pair.
right). Light was launched from these fibers to
facilitate the alignment.
CCD #2
Positive lens
Lenslet array
Negative lens
Output V -groove
fiber array
CCD #1
Figure 8. Auxiliary components utilized to align the output V-groove array of fibers. A pellicle beam splitter is
used to image the lenslet plane onto a TV camera (CCD #1). A CCD array (CCD #2), which is conjugate to the
leuslet, detects the focusing spot reflected by the pellicle.
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
Output detector #2
Input laser #9
Input laser #2
Output detector #2
Figure 9. Input modulation waveforms applied to two of the laser diodes and corresponding detected outputs. The
presence of crosstalk is quite severe but can be minimized with optimal alignment.
4.2. Switching experiments
The AO switch was configured to set up a particular interconnection pattern while the lasers were pulsed with
easily distinguishable waveforms. Then the RF multitone waveform was switched to a new multitone superposition
representing a new interconnection matrix. This allowed the measurement of the transition time to be under 20 s.
Examples of output detector waveforms and the modulation waveforms are shown in Fig. 9. Feedback from the
fiber facets made it difficult to obtain clean laser modulations, and the outputs are also corrupted with substantial
crosstalk. The laser operation can be improved using angle polished connectors to minimize feedback. The crosstalk
can be minimized (to better than 25 dB) with improved alignment.
A 12x12 acousto-optic photonic crossbar that allows rapidly reconfigurable interconnections between single mode
fibers has been demonstrated. The system was designed to operate as a low-loss diffraction limited interconnect
between 12 single mode 5 jm input fiber cores to 12 single mode 5 um output fiber cores, but the alignment to
achieve this performance proved challenging and the procedure was presented. The system is fully controllable from
a computer interface and this flexibility was utilized to investigate the reconfiguration speed, which is limited by the
acoustic transit time across the device aperture, and was shown to be less than 20 ,as
This work was partly funded by the NSF Young Investigator program ECS 9258088, and by the Spanish Ministry of
Education through the CICYT program.
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
1. R. Mcleod, K. Y. Wu, K. Wagner, and R. T. Weverka, "Acoustooptic photonic crossbar switch: Part 1 - design,"
Applied Optics 35(32), pp. 6331—6353, 1996.
2. R. Mcleod, PhD dissertation, University of Colorado, Boulder, CO, 1995.
3. R. M. K. W. R. T. Weverka, K. Wagner and C. Garvin, Acousto-Optic Signal Proccesing: Theory and Implementation, ch. Low-loss acousto-optic photonic switch, pp. 479—573. Marcel Dekker Inc., New York, 2nd. ed.,
4. B. Mukherjee, "WDM-based local lightwave networks. part II: Multihop systems," IEEE Network 6(4), p. 20,
5. Sinclair Optics, Inc., OSLO Series 2 and 3 Operating Manual, first ed., 1991.
6. D. C. O'Shea, Elements of Modern Optical Design, John Wiley and Sons, New York, 1985.
Downloaded from SPIE Digital Library on 07 Jun 2012 to Terms of Use: http://spiedl.org/terms
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF